xref: /openbmc/linux/net/sunrpc/xprtsock.c (revision 3c341b0b925eee01daae2c594b81e673f659d7cd)
1 /*
2  * linux/net/sunrpc/xprtsock.c
3  *
4  * Client-side transport implementation for sockets.
5  *
6  * TCP callback races fixes (C) 1998 Red Hat Software <alan@redhat.com>
7  * TCP send fixes (C) 1998 Red Hat Software <alan@redhat.com>
8  * TCP NFS related read + write fixes
9  *  (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie>
10  *
11  * Rewrite of larges part of the code in order to stabilize TCP stuff.
12  * Fix behaviour when socket buffer is full.
13  *  (C) 1999 Trond Myklebust <trond.myklebust@fys.uio.no>
14  *
15  * IP socket transport implementation, (C) 2005 Chuck Lever <cel@netapp.com>
16  *
17  * IPv6 support contributed by Gilles Quillard, Bull Open Source, 2005.
18  *   <gilles.quillard@bull.net>
19  */
20 
21 #include <linux/types.h>
22 #include <linux/slab.h>
23 #include <linux/module.h>
24 #include <linux/capability.h>
25 #include <linux/pagemap.h>
26 #include <linux/errno.h>
27 #include <linux/socket.h>
28 #include <linux/in.h>
29 #include <linux/net.h>
30 #include <linux/mm.h>
31 #include <linux/udp.h>
32 #include <linux/tcp.h>
33 #include <linux/sunrpc/clnt.h>
34 #include <linux/sunrpc/sched.h>
35 #include <linux/file.h>
36 
37 #include <net/sock.h>
38 #include <net/checksum.h>
39 #include <net/udp.h>
40 #include <net/tcp.h>
41 
42 /*
43  * xprtsock tunables
44  */
45 unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
46 unsigned int xprt_tcp_slot_table_entries = RPC_DEF_SLOT_TABLE;
47 
48 unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
49 unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
50 
51 /*
52  * We can register our own files under /proc/sys/sunrpc by
53  * calling register_sysctl_table() again.  The files in that
54  * directory become the union of all files registered there.
55  *
56  * We simply need to make sure that we don't collide with
57  * someone else's file names!
58  */
59 
60 #ifdef RPC_DEBUG
61 
62 static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
63 static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
64 static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
65 static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
66 
67 static struct ctl_table_header *sunrpc_table_header;
68 
69 /*
70  * FIXME: changing the UDP slot table size should also resize the UDP
71  *        socket buffers for existing UDP transports
72  */
73 static ctl_table xs_tunables_table[] = {
74 	{
75 		.ctl_name	= CTL_SLOTTABLE_UDP,
76 		.procname	= "udp_slot_table_entries",
77 		.data		= &xprt_udp_slot_table_entries,
78 		.maxlen		= sizeof(unsigned int),
79 		.mode		= 0644,
80 		.proc_handler	= &proc_dointvec_minmax,
81 		.strategy	= &sysctl_intvec,
82 		.extra1		= &min_slot_table_size,
83 		.extra2		= &max_slot_table_size
84 	},
85 	{
86 		.ctl_name	= CTL_SLOTTABLE_TCP,
87 		.procname	= "tcp_slot_table_entries",
88 		.data		= &xprt_tcp_slot_table_entries,
89 		.maxlen		= sizeof(unsigned int),
90 		.mode		= 0644,
91 		.proc_handler	= &proc_dointvec_minmax,
92 		.strategy	= &sysctl_intvec,
93 		.extra1		= &min_slot_table_size,
94 		.extra2		= &max_slot_table_size
95 	},
96 	{
97 		.ctl_name	= CTL_MIN_RESVPORT,
98 		.procname	= "min_resvport",
99 		.data		= &xprt_min_resvport,
100 		.maxlen		= sizeof(unsigned int),
101 		.mode		= 0644,
102 		.proc_handler	= &proc_dointvec_minmax,
103 		.strategy	= &sysctl_intvec,
104 		.extra1		= &xprt_min_resvport_limit,
105 		.extra2		= &xprt_max_resvport_limit
106 	},
107 	{
108 		.ctl_name	= CTL_MAX_RESVPORT,
109 		.procname	= "max_resvport",
110 		.data		= &xprt_max_resvport,
111 		.maxlen		= sizeof(unsigned int),
112 		.mode		= 0644,
113 		.proc_handler	= &proc_dointvec_minmax,
114 		.strategy	= &sysctl_intvec,
115 		.extra1		= &xprt_min_resvport_limit,
116 		.extra2		= &xprt_max_resvport_limit
117 	},
118 	{
119 		.ctl_name = 0,
120 	},
121 };
122 
123 static ctl_table sunrpc_table[] = {
124 	{
125 		.ctl_name	= CTL_SUNRPC,
126 		.procname	= "sunrpc",
127 		.mode		= 0555,
128 		.child		= xs_tunables_table
129 	},
130 	{
131 		.ctl_name = 0,
132 	},
133 };
134 
135 #endif
136 
137 /*
138  * How many times to try sending a request on a socket before waiting
139  * for the socket buffer to clear.
140  */
141 #define XS_SENDMSG_RETRY	(10U)
142 
143 /*
144  * Time out for an RPC UDP socket connect.  UDP socket connects are
145  * synchronous, but we set a timeout anyway in case of resource
146  * exhaustion on the local host.
147  */
148 #define XS_UDP_CONN_TO		(5U * HZ)
149 
150 /*
151  * Wait duration for an RPC TCP connection to be established.  Solaris
152  * NFS over TCP uses 60 seconds, for example, which is in line with how
153  * long a server takes to reboot.
154  */
155 #define XS_TCP_CONN_TO		(60U * HZ)
156 
157 /*
158  * Wait duration for a reply from the RPC portmapper.
159  */
160 #define XS_BIND_TO		(60U * HZ)
161 
162 /*
163  * Delay if a UDP socket connect error occurs.  This is most likely some
164  * kind of resource problem on the local host.
165  */
166 #define XS_UDP_REEST_TO		(2U * HZ)
167 
168 /*
169  * The reestablish timeout allows clients to delay for a bit before attempting
170  * to reconnect to a server that just dropped our connection.
171  *
172  * We implement an exponential backoff when trying to reestablish a TCP
173  * transport connection with the server.  Some servers like to drop a TCP
174  * connection when they are overworked, so we start with a short timeout and
175  * increase over time if the server is down or not responding.
176  */
177 #define XS_TCP_INIT_REEST_TO	(3U * HZ)
178 #define XS_TCP_MAX_REEST_TO	(5U * 60 * HZ)
179 
180 /*
181  * TCP idle timeout; client drops the transport socket if it is idle
182  * for this long.  Note that we also timeout UDP sockets to prevent
183  * holding port numbers when there is no RPC traffic.
184  */
185 #define XS_IDLE_DISC_TO		(5U * 60 * HZ)
186 
187 #ifdef RPC_DEBUG
188 # undef  RPC_DEBUG_DATA
189 # define RPCDBG_FACILITY	RPCDBG_TRANS
190 #endif
191 
192 #ifdef RPC_DEBUG_DATA
193 static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
194 {
195 	u8 *buf = (u8 *) packet;
196 	int j;
197 
198 	dprintk("RPC:       %s\n", msg);
199 	for (j = 0; j < count && j < 128; j += 4) {
200 		if (!(j & 31)) {
201 			if (j)
202 				dprintk("\n");
203 			dprintk("0x%04x ", j);
204 		}
205 		dprintk("%02x%02x%02x%02x ",
206 			buf[j], buf[j+1], buf[j+2], buf[j+3]);
207 	}
208 	dprintk("\n");
209 }
210 #else
211 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
212 {
213 	/* NOP */
214 }
215 #endif
216 
217 struct sock_xprt {
218 	struct rpc_xprt		xprt;
219 
220 	/*
221 	 * Network layer
222 	 */
223 	struct socket *		sock;
224 	struct sock *		inet;
225 
226 	/*
227 	 * State of TCP reply receive
228 	 */
229 	__be32			tcp_fraghdr,
230 				tcp_xid;
231 
232 	u32			tcp_offset,
233 				tcp_reclen;
234 
235 	unsigned long		tcp_copied,
236 				tcp_flags;
237 
238 	/*
239 	 * Connection of transports
240 	 */
241 	struct delayed_work	connect_worker;
242 	struct sockaddr_storage	addr;
243 	unsigned short		port;
244 
245 	/*
246 	 * UDP socket buffer size parameters
247 	 */
248 	size_t			rcvsize,
249 				sndsize;
250 
251 	/*
252 	 * Saved socket callback addresses
253 	 */
254 	void			(*old_data_ready)(struct sock *, int);
255 	void			(*old_state_change)(struct sock *);
256 	void			(*old_write_space)(struct sock *);
257 };
258 
259 /*
260  * TCP receive state flags
261  */
262 #define TCP_RCV_LAST_FRAG	(1UL << 0)
263 #define TCP_RCV_COPY_FRAGHDR	(1UL << 1)
264 #define TCP_RCV_COPY_XID	(1UL << 2)
265 #define TCP_RCV_COPY_DATA	(1UL << 3)
266 
267 static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
268 {
269 	return (struct sockaddr *) &xprt->addr;
270 }
271 
272 static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
273 {
274 	return (struct sockaddr_in *) &xprt->addr;
275 }
276 
277 static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
278 {
279 	return (struct sockaddr_in6 *) &xprt->addr;
280 }
281 
282 static void xs_format_ipv4_peer_addresses(struct rpc_xprt *xprt)
283 {
284 	struct sockaddr_in *addr = xs_addr_in(xprt);
285 	char *buf;
286 
287 	buf = kzalloc(20, GFP_KERNEL);
288 	if (buf) {
289 		snprintf(buf, 20, NIPQUAD_FMT,
290 				NIPQUAD(addr->sin_addr.s_addr));
291 	}
292 	xprt->address_strings[RPC_DISPLAY_ADDR] = buf;
293 
294 	buf = kzalloc(8, GFP_KERNEL);
295 	if (buf) {
296 		snprintf(buf, 8, "%u",
297 				ntohs(addr->sin_port));
298 	}
299 	xprt->address_strings[RPC_DISPLAY_PORT] = buf;
300 
301 	buf = kzalloc(8, GFP_KERNEL);
302 	if (buf) {
303 		if (xprt->prot == IPPROTO_UDP)
304 			snprintf(buf, 8, "udp");
305 		else
306 			snprintf(buf, 8, "tcp");
307 	}
308 	xprt->address_strings[RPC_DISPLAY_PROTO] = buf;
309 
310 	buf = kzalloc(48, GFP_KERNEL);
311 	if (buf) {
312 		snprintf(buf, 48, "addr="NIPQUAD_FMT" port=%u proto=%s",
313 			NIPQUAD(addr->sin_addr.s_addr),
314 			ntohs(addr->sin_port),
315 			xprt->prot == IPPROTO_UDP ? "udp" : "tcp");
316 	}
317 	xprt->address_strings[RPC_DISPLAY_ALL] = buf;
318 
319 	buf = kzalloc(10, GFP_KERNEL);
320 	if (buf) {
321 		snprintf(buf, 10, "%02x%02x%02x%02x",
322 				NIPQUAD(addr->sin_addr.s_addr));
323 	}
324 	xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = buf;
325 
326 	buf = kzalloc(8, GFP_KERNEL);
327 	if (buf) {
328 		snprintf(buf, 8, "%4hx",
329 				ntohs(addr->sin_port));
330 	}
331 	xprt->address_strings[RPC_DISPLAY_HEX_PORT] = buf;
332 
333 	buf = kzalloc(30, GFP_KERNEL);
334 	if (buf) {
335 		snprintf(buf, 30, NIPQUAD_FMT".%u.%u",
336 				NIPQUAD(addr->sin_addr.s_addr),
337 				ntohs(addr->sin_port) >> 8,
338 				ntohs(addr->sin_port) & 0xff);
339 	}
340 	xprt->address_strings[RPC_DISPLAY_UNIVERSAL_ADDR] = buf;
341 
342 	xprt->address_strings[RPC_DISPLAY_NETID] =
343 		kstrdup(xprt->prot == IPPROTO_UDP ?
344 			RPCBIND_NETID_UDP : RPCBIND_NETID_TCP, GFP_KERNEL);
345 }
346 
347 static void xs_format_ipv6_peer_addresses(struct rpc_xprt *xprt)
348 {
349 	struct sockaddr_in6 *addr = xs_addr_in6(xprt);
350 	char *buf;
351 
352 	buf = kzalloc(40, GFP_KERNEL);
353 	if (buf) {
354 		snprintf(buf, 40, NIP6_FMT,
355 				NIP6(addr->sin6_addr));
356 	}
357 	xprt->address_strings[RPC_DISPLAY_ADDR] = buf;
358 
359 	buf = kzalloc(8, GFP_KERNEL);
360 	if (buf) {
361 		snprintf(buf, 8, "%u",
362 				ntohs(addr->sin6_port));
363 	}
364 	xprt->address_strings[RPC_DISPLAY_PORT] = buf;
365 
366 	buf = kzalloc(8, GFP_KERNEL);
367 	if (buf) {
368 		if (xprt->prot == IPPROTO_UDP)
369 			snprintf(buf, 8, "udp");
370 		else
371 			snprintf(buf, 8, "tcp");
372 	}
373 	xprt->address_strings[RPC_DISPLAY_PROTO] = buf;
374 
375 	buf = kzalloc(64, GFP_KERNEL);
376 	if (buf) {
377 		snprintf(buf, 64, "addr="NIP6_FMT" port=%u proto=%s",
378 				NIP6(addr->sin6_addr),
379 				ntohs(addr->sin6_port),
380 				xprt->prot == IPPROTO_UDP ? "udp" : "tcp");
381 	}
382 	xprt->address_strings[RPC_DISPLAY_ALL] = buf;
383 
384 	buf = kzalloc(36, GFP_KERNEL);
385 	if (buf) {
386 		snprintf(buf, 36, NIP6_SEQFMT,
387 				NIP6(addr->sin6_addr));
388 	}
389 	xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = buf;
390 
391 	buf = kzalloc(8, GFP_KERNEL);
392 	if (buf) {
393 		snprintf(buf, 8, "%4hx",
394 				ntohs(addr->sin6_port));
395 	}
396 	xprt->address_strings[RPC_DISPLAY_HEX_PORT] = buf;
397 
398 	buf = kzalloc(50, GFP_KERNEL);
399 	if (buf) {
400 		snprintf(buf, 50, NIP6_FMT".%u.%u",
401 				NIP6(addr->sin6_addr),
402 				ntohs(addr->sin6_port) >> 8,
403 				ntohs(addr->sin6_port) & 0xff);
404 	}
405 	xprt->address_strings[RPC_DISPLAY_UNIVERSAL_ADDR] = buf;
406 
407 	xprt->address_strings[RPC_DISPLAY_NETID] =
408 		kstrdup(xprt->prot == IPPROTO_UDP ?
409 			RPCBIND_NETID_UDP6 : RPCBIND_NETID_TCP6, GFP_KERNEL);
410 }
411 
412 static void xs_free_peer_addresses(struct rpc_xprt *xprt)
413 {
414 	int i;
415 
416 	for (i = 0; i < RPC_DISPLAY_MAX; i++)
417 		kfree(xprt->address_strings[i]);
418 }
419 
420 #define XS_SENDMSG_FLAGS	(MSG_DONTWAIT | MSG_NOSIGNAL)
421 
422 static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
423 {
424 	struct msghdr msg = {
425 		.msg_name	= addr,
426 		.msg_namelen	= addrlen,
427 		.msg_flags	= XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
428 	};
429 	struct kvec iov = {
430 		.iov_base	= vec->iov_base + base,
431 		.iov_len	= vec->iov_len - base,
432 	};
433 
434 	if (iov.iov_len != 0)
435 		return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
436 	return kernel_sendmsg(sock, &msg, NULL, 0, 0);
437 }
438 
439 static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more)
440 {
441 	struct page **ppage;
442 	unsigned int remainder;
443 	int err, sent = 0;
444 
445 	remainder = xdr->page_len - base;
446 	base += xdr->page_base;
447 	ppage = xdr->pages + (base >> PAGE_SHIFT);
448 	base &= ~PAGE_MASK;
449 	for(;;) {
450 		unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
451 		int flags = XS_SENDMSG_FLAGS;
452 
453 		remainder -= len;
454 		if (remainder != 0 || more)
455 			flags |= MSG_MORE;
456 		err = sock->ops->sendpage(sock, *ppage, base, len, flags);
457 		if (remainder == 0 || err != len)
458 			break;
459 		sent += err;
460 		ppage++;
461 		base = 0;
462 	}
463 	if (sent == 0)
464 		return err;
465 	if (err > 0)
466 		sent += err;
467 	return sent;
468 }
469 
470 /**
471  * xs_sendpages - write pages directly to a socket
472  * @sock: socket to send on
473  * @addr: UDP only -- address of destination
474  * @addrlen: UDP only -- length of destination address
475  * @xdr: buffer containing this request
476  * @base: starting position in the buffer
477  *
478  */
479 static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base)
480 {
481 	unsigned int remainder = xdr->len - base;
482 	int err, sent = 0;
483 
484 	if (unlikely(!sock))
485 		return -ENOTCONN;
486 
487 	clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
488 	if (base != 0) {
489 		addr = NULL;
490 		addrlen = 0;
491 	}
492 
493 	if (base < xdr->head[0].iov_len || addr != NULL) {
494 		unsigned int len = xdr->head[0].iov_len - base;
495 		remainder -= len;
496 		err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
497 		if (remainder == 0 || err != len)
498 			goto out;
499 		sent += err;
500 		base = 0;
501 	} else
502 		base -= xdr->head[0].iov_len;
503 
504 	if (base < xdr->page_len) {
505 		unsigned int len = xdr->page_len - base;
506 		remainder -= len;
507 		err = xs_send_pagedata(sock, xdr, base, remainder != 0);
508 		if (remainder == 0 || err != len)
509 			goto out;
510 		sent += err;
511 		base = 0;
512 	} else
513 		base -= xdr->page_len;
514 
515 	if (base >= xdr->tail[0].iov_len)
516 		return sent;
517 	err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
518 out:
519 	if (sent == 0)
520 		return err;
521 	if (err > 0)
522 		sent += err;
523 	return sent;
524 }
525 
526 /**
527  * xs_nospace - place task on wait queue if transmit was incomplete
528  * @task: task to put to sleep
529  *
530  */
531 static void xs_nospace(struct rpc_task *task)
532 {
533 	struct rpc_rqst *req = task->tk_rqstp;
534 	struct rpc_xprt *xprt = req->rq_xprt;
535 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
536 
537 	dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
538 			task->tk_pid, req->rq_slen - req->rq_bytes_sent,
539 			req->rq_slen);
540 
541 	if (test_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags)) {
542 		/* Protect against races with write_space */
543 		spin_lock_bh(&xprt->transport_lock);
544 
545 		/* Don't race with disconnect */
546 		if (!xprt_connected(xprt))
547 			task->tk_status = -ENOTCONN;
548 		else if (test_bit(SOCK_NOSPACE, &transport->sock->flags))
549 			xprt_wait_for_buffer_space(task);
550 
551 		spin_unlock_bh(&xprt->transport_lock);
552 	} else
553 		/* Keep holding the socket if it is blocked */
554 		rpc_delay(task, HZ>>4);
555 }
556 
557 /**
558  * xs_udp_send_request - write an RPC request to a UDP socket
559  * @task: address of RPC task that manages the state of an RPC request
560  *
561  * Return values:
562  *        0:	The request has been sent
563  *   EAGAIN:	The socket was blocked, please call again later to
564  *		complete the request
565  * ENOTCONN:	Caller needs to invoke connect logic then call again
566  *    other:	Some other error occured, the request was not sent
567  */
568 static int xs_udp_send_request(struct rpc_task *task)
569 {
570 	struct rpc_rqst *req = task->tk_rqstp;
571 	struct rpc_xprt *xprt = req->rq_xprt;
572 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
573 	struct xdr_buf *xdr = &req->rq_snd_buf;
574 	int status;
575 
576 	xs_pktdump("packet data:",
577 				req->rq_svec->iov_base,
578 				req->rq_svec->iov_len);
579 
580 	req->rq_xtime = jiffies;
581 	status = xs_sendpages(transport->sock,
582 			      xs_addr(xprt),
583 			      xprt->addrlen, xdr,
584 			      req->rq_bytes_sent);
585 
586 	dprintk("RPC:       xs_udp_send_request(%u) = %d\n",
587 			xdr->len - req->rq_bytes_sent, status);
588 
589 	if (likely(status >= (int) req->rq_slen))
590 		return 0;
591 
592 	/* Still some bytes left; set up for a retry later. */
593 	if (status > 0)
594 		status = -EAGAIN;
595 
596 	switch (status) {
597 	case -ENETUNREACH:
598 	case -EPIPE:
599 	case -ECONNREFUSED:
600 		/* When the server has died, an ICMP port unreachable message
601 		 * prompts ECONNREFUSED. */
602 		break;
603 	case -EAGAIN:
604 		xs_nospace(task);
605 		break;
606 	default:
607 		dprintk("RPC:       sendmsg returned unrecognized error %d\n",
608 			-status);
609 		break;
610 	}
611 
612 	return status;
613 }
614 
615 static inline void xs_encode_tcp_record_marker(struct xdr_buf *buf)
616 {
617 	u32 reclen = buf->len - sizeof(rpc_fraghdr);
618 	rpc_fraghdr *base = buf->head[0].iov_base;
619 	*base = htonl(RPC_LAST_STREAM_FRAGMENT | reclen);
620 }
621 
622 /**
623  * xs_tcp_send_request - write an RPC request to a TCP socket
624  * @task: address of RPC task that manages the state of an RPC request
625  *
626  * Return values:
627  *        0:	The request has been sent
628  *   EAGAIN:	The socket was blocked, please call again later to
629  *		complete the request
630  * ENOTCONN:	Caller needs to invoke connect logic then call again
631  *    other:	Some other error occured, the request was not sent
632  *
633  * XXX: In the case of soft timeouts, should we eventually give up
634  *	if sendmsg is not able to make progress?
635  */
636 static int xs_tcp_send_request(struct rpc_task *task)
637 {
638 	struct rpc_rqst *req = task->tk_rqstp;
639 	struct rpc_xprt *xprt = req->rq_xprt;
640 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
641 	struct xdr_buf *xdr = &req->rq_snd_buf;
642 	int status;
643 	unsigned int retry = 0;
644 
645 	xs_encode_tcp_record_marker(&req->rq_snd_buf);
646 
647 	xs_pktdump("packet data:",
648 				req->rq_svec->iov_base,
649 				req->rq_svec->iov_len);
650 
651 	/* Continue transmitting the packet/record. We must be careful
652 	 * to cope with writespace callbacks arriving _after_ we have
653 	 * called sendmsg(). */
654 	while (1) {
655 		req->rq_xtime = jiffies;
656 		status = xs_sendpages(transport->sock,
657 					NULL, 0, xdr, req->rq_bytes_sent);
658 
659 		dprintk("RPC:       xs_tcp_send_request(%u) = %d\n",
660 				xdr->len - req->rq_bytes_sent, status);
661 
662 		if (unlikely(status < 0))
663 			break;
664 
665 		/* If we've sent the entire packet, immediately
666 		 * reset the count of bytes sent. */
667 		req->rq_bytes_sent += status;
668 		task->tk_bytes_sent += status;
669 		if (likely(req->rq_bytes_sent >= req->rq_slen)) {
670 			req->rq_bytes_sent = 0;
671 			return 0;
672 		}
673 
674 		status = -EAGAIN;
675 		if (retry++ > XS_SENDMSG_RETRY)
676 			break;
677 	}
678 
679 	switch (status) {
680 	case -EAGAIN:
681 		xs_nospace(task);
682 		break;
683 	case -ECONNREFUSED:
684 	case -ECONNRESET:
685 	case -ENOTCONN:
686 	case -EPIPE:
687 		status = -ENOTCONN;
688 		break;
689 	default:
690 		dprintk("RPC:       sendmsg returned unrecognized error %d\n",
691 			-status);
692 		xprt_disconnect(xprt);
693 		break;
694 	}
695 
696 	return status;
697 }
698 
699 /**
700  * xs_tcp_release_xprt - clean up after a tcp transmission
701  * @xprt: transport
702  * @task: rpc task
703  *
704  * This cleans up if an error causes us to abort the transmission of a request.
705  * In this case, the socket may need to be reset in order to avoid confusing
706  * the server.
707  */
708 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
709 {
710 	struct rpc_rqst *req;
711 
712 	if (task != xprt->snd_task)
713 		return;
714 	if (task == NULL)
715 		goto out_release;
716 	req = task->tk_rqstp;
717 	if (req->rq_bytes_sent == 0)
718 		goto out_release;
719 	if (req->rq_bytes_sent == req->rq_snd_buf.len)
720 		goto out_release;
721 	set_bit(XPRT_CLOSE_WAIT, &task->tk_xprt->state);
722 out_release:
723 	xprt_release_xprt(xprt, task);
724 }
725 
726 /**
727  * xs_close - close a socket
728  * @xprt: transport
729  *
730  * This is used when all requests are complete; ie, no DRC state remains
731  * on the server we want to save.
732  */
733 static void xs_close(struct rpc_xprt *xprt)
734 {
735 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
736 	struct socket *sock = transport->sock;
737 	struct sock *sk = transport->inet;
738 
739 	if (!sk)
740 		goto clear_close_wait;
741 
742 	dprintk("RPC:       xs_close xprt %p\n", xprt);
743 
744 	write_lock_bh(&sk->sk_callback_lock);
745 	transport->inet = NULL;
746 	transport->sock = NULL;
747 
748 	sk->sk_user_data = NULL;
749 	sk->sk_data_ready = transport->old_data_ready;
750 	sk->sk_state_change = transport->old_state_change;
751 	sk->sk_write_space = transport->old_write_space;
752 	write_unlock_bh(&sk->sk_callback_lock);
753 
754 	sk->sk_no_check = 0;
755 
756 	sock_release(sock);
757 clear_close_wait:
758 	smp_mb__before_clear_bit();
759 	clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
760 	smp_mb__after_clear_bit();
761 }
762 
763 /**
764  * xs_destroy - prepare to shutdown a transport
765  * @xprt: doomed transport
766  *
767  */
768 static void xs_destroy(struct rpc_xprt *xprt)
769 {
770 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
771 
772 	dprintk("RPC:       xs_destroy xprt %p\n", xprt);
773 
774 	cancel_rearming_delayed_work(&transport->connect_worker);
775 
776 	xprt_disconnect(xprt);
777 	xs_close(xprt);
778 	xs_free_peer_addresses(xprt);
779 	kfree(xprt->slot);
780 	kfree(xprt);
781 	module_put(THIS_MODULE);
782 }
783 
784 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
785 {
786 	return (struct rpc_xprt *) sk->sk_user_data;
787 }
788 
789 /**
790  * xs_udp_data_ready - "data ready" callback for UDP sockets
791  * @sk: socket with data to read
792  * @len: how much data to read
793  *
794  */
795 static void xs_udp_data_ready(struct sock *sk, int len)
796 {
797 	struct rpc_task *task;
798 	struct rpc_xprt *xprt;
799 	struct rpc_rqst *rovr;
800 	struct sk_buff *skb;
801 	int err, repsize, copied;
802 	u32 _xid;
803 	__be32 *xp;
804 
805 	read_lock(&sk->sk_callback_lock);
806 	dprintk("RPC:       xs_udp_data_ready...\n");
807 	if (!(xprt = xprt_from_sock(sk)))
808 		goto out;
809 
810 	if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
811 		goto out;
812 
813 	if (xprt->shutdown)
814 		goto dropit;
815 
816 	repsize = skb->len - sizeof(struct udphdr);
817 	if (repsize < 4) {
818 		dprintk("RPC:       impossible RPC reply size %d!\n", repsize);
819 		goto dropit;
820 	}
821 
822 	/* Copy the XID from the skb... */
823 	xp = skb_header_pointer(skb, sizeof(struct udphdr),
824 				sizeof(_xid), &_xid);
825 	if (xp == NULL)
826 		goto dropit;
827 
828 	/* Look up and lock the request corresponding to the given XID */
829 	spin_lock(&xprt->transport_lock);
830 	rovr = xprt_lookup_rqst(xprt, *xp);
831 	if (!rovr)
832 		goto out_unlock;
833 	task = rovr->rq_task;
834 
835 	if ((copied = rovr->rq_private_buf.buflen) > repsize)
836 		copied = repsize;
837 
838 	/* Suck it into the iovec, verify checksum if not done by hw. */
839 	if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb))
840 		goto out_unlock;
841 
842 	/* Something worked... */
843 	dst_confirm(skb->dst);
844 
845 	xprt_adjust_cwnd(task, copied);
846 	xprt_update_rtt(task);
847 	xprt_complete_rqst(task, copied);
848 
849  out_unlock:
850 	spin_unlock(&xprt->transport_lock);
851  dropit:
852 	skb_free_datagram(sk, skb);
853  out:
854 	read_unlock(&sk->sk_callback_lock);
855 }
856 
857 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
858 {
859 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
860 	size_t len, used;
861 	char *p;
862 
863 	p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
864 	len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
865 	used = xdr_skb_read_bits(desc, p, len);
866 	transport->tcp_offset += used;
867 	if (used != len)
868 		return;
869 
870 	transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
871 	if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
872 		transport->tcp_flags |= TCP_RCV_LAST_FRAG;
873 	else
874 		transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
875 	transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
876 
877 	transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
878 	transport->tcp_offset = 0;
879 
880 	/* Sanity check of the record length */
881 	if (unlikely(transport->tcp_reclen < 4)) {
882 		dprintk("RPC:       invalid TCP record fragment length\n");
883 		xprt_disconnect(xprt);
884 		return;
885 	}
886 	dprintk("RPC:       reading TCP record fragment of length %d\n",
887 			transport->tcp_reclen);
888 }
889 
890 static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
891 {
892 	if (transport->tcp_offset == transport->tcp_reclen) {
893 		transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
894 		transport->tcp_offset = 0;
895 		if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
896 			transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
897 			transport->tcp_flags |= TCP_RCV_COPY_XID;
898 			transport->tcp_copied = 0;
899 		}
900 	}
901 }
902 
903 static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
904 {
905 	size_t len, used;
906 	char *p;
907 
908 	len = sizeof(transport->tcp_xid) - transport->tcp_offset;
909 	dprintk("RPC:       reading XID (%Zu bytes)\n", len);
910 	p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
911 	used = xdr_skb_read_bits(desc, p, len);
912 	transport->tcp_offset += used;
913 	if (used != len)
914 		return;
915 	transport->tcp_flags &= ~TCP_RCV_COPY_XID;
916 	transport->tcp_flags |= TCP_RCV_COPY_DATA;
917 	transport->tcp_copied = 4;
918 	dprintk("RPC:       reading reply for XID %08x\n",
919 			ntohl(transport->tcp_xid));
920 	xs_tcp_check_fraghdr(transport);
921 }
922 
923 static inline void xs_tcp_read_request(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
924 {
925 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
926 	struct rpc_rqst *req;
927 	struct xdr_buf *rcvbuf;
928 	size_t len;
929 	ssize_t r;
930 
931 	/* Find and lock the request corresponding to this xid */
932 	spin_lock(&xprt->transport_lock);
933 	req = xprt_lookup_rqst(xprt, transport->tcp_xid);
934 	if (!req) {
935 		transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
936 		dprintk("RPC:       XID %08x request not found!\n",
937 				ntohl(transport->tcp_xid));
938 		spin_unlock(&xprt->transport_lock);
939 		return;
940 	}
941 
942 	rcvbuf = &req->rq_private_buf;
943 	len = desc->count;
944 	if (len > transport->tcp_reclen - transport->tcp_offset) {
945 		struct xdr_skb_reader my_desc;
946 
947 		len = transport->tcp_reclen - transport->tcp_offset;
948 		memcpy(&my_desc, desc, sizeof(my_desc));
949 		my_desc.count = len;
950 		r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
951 					  &my_desc, xdr_skb_read_bits);
952 		desc->count -= r;
953 		desc->offset += r;
954 	} else
955 		r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
956 					  desc, xdr_skb_read_bits);
957 
958 	if (r > 0) {
959 		transport->tcp_copied += r;
960 		transport->tcp_offset += r;
961 	}
962 	if (r != len) {
963 		/* Error when copying to the receive buffer,
964 		 * usually because we weren't able to allocate
965 		 * additional buffer pages. All we can do now
966 		 * is turn off TCP_RCV_COPY_DATA, so the request
967 		 * will not receive any additional updates,
968 		 * and time out.
969 		 * Any remaining data from this record will
970 		 * be discarded.
971 		 */
972 		transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
973 		dprintk("RPC:       XID %08x truncated request\n",
974 				ntohl(transport->tcp_xid));
975 		dprintk("RPC:       xprt = %p, tcp_copied = %lu, "
976 				"tcp_offset = %u, tcp_reclen = %u\n",
977 				xprt, transport->tcp_copied,
978 				transport->tcp_offset, transport->tcp_reclen);
979 		goto out;
980 	}
981 
982 	dprintk("RPC:       XID %08x read %Zd bytes\n",
983 			ntohl(transport->tcp_xid), r);
984 	dprintk("RPC:       xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
985 			"tcp_reclen = %u\n", xprt, transport->tcp_copied,
986 			transport->tcp_offset, transport->tcp_reclen);
987 
988 	if (transport->tcp_copied == req->rq_private_buf.buflen)
989 		transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
990 	else if (transport->tcp_offset == transport->tcp_reclen) {
991 		if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
992 			transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
993 	}
994 
995 out:
996 	if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
997 		xprt_complete_rqst(req->rq_task, transport->tcp_copied);
998 	spin_unlock(&xprt->transport_lock);
999 	xs_tcp_check_fraghdr(transport);
1000 }
1001 
1002 static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1003 {
1004 	size_t len;
1005 
1006 	len = transport->tcp_reclen - transport->tcp_offset;
1007 	if (len > desc->count)
1008 		len = desc->count;
1009 	desc->count -= len;
1010 	desc->offset += len;
1011 	transport->tcp_offset += len;
1012 	dprintk("RPC:       discarded %Zu bytes\n", len);
1013 	xs_tcp_check_fraghdr(transport);
1014 }
1015 
1016 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
1017 {
1018 	struct rpc_xprt *xprt = rd_desc->arg.data;
1019 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1020 	struct xdr_skb_reader desc = {
1021 		.skb	= skb,
1022 		.offset	= offset,
1023 		.count	= len,
1024 	};
1025 
1026 	dprintk("RPC:       xs_tcp_data_recv started\n");
1027 	do {
1028 		/* Read in a new fragment marker if necessary */
1029 		/* Can we ever really expect to get completely empty fragments? */
1030 		if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
1031 			xs_tcp_read_fraghdr(xprt, &desc);
1032 			continue;
1033 		}
1034 		/* Read in the xid if necessary */
1035 		if (transport->tcp_flags & TCP_RCV_COPY_XID) {
1036 			xs_tcp_read_xid(transport, &desc);
1037 			continue;
1038 		}
1039 		/* Read in the request data */
1040 		if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
1041 			xs_tcp_read_request(xprt, &desc);
1042 			continue;
1043 		}
1044 		/* Skip over any trailing bytes on short reads */
1045 		xs_tcp_read_discard(transport, &desc);
1046 	} while (desc.count);
1047 	dprintk("RPC:       xs_tcp_data_recv done\n");
1048 	return len - desc.count;
1049 }
1050 
1051 /**
1052  * xs_tcp_data_ready - "data ready" callback for TCP sockets
1053  * @sk: socket with data to read
1054  * @bytes: how much data to read
1055  *
1056  */
1057 static void xs_tcp_data_ready(struct sock *sk, int bytes)
1058 {
1059 	struct rpc_xprt *xprt;
1060 	read_descriptor_t rd_desc;
1061 
1062 	dprintk("RPC:       xs_tcp_data_ready...\n");
1063 
1064 	read_lock(&sk->sk_callback_lock);
1065 	if (!(xprt = xprt_from_sock(sk)))
1066 		goto out;
1067 	if (xprt->shutdown)
1068 		goto out;
1069 
1070 	/* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1071 	rd_desc.arg.data = xprt;
1072 	rd_desc.count = 65536;
1073 	tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1074 out:
1075 	read_unlock(&sk->sk_callback_lock);
1076 }
1077 
1078 /**
1079  * xs_tcp_state_change - callback to handle TCP socket state changes
1080  * @sk: socket whose state has changed
1081  *
1082  */
1083 static void xs_tcp_state_change(struct sock *sk)
1084 {
1085 	struct rpc_xprt *xprt;
1086 
1087 	read_lock(&sk->sk_callback_lock);
1088 	if (!(xprt = xprt_from_sock(sk)))
1089 		goto out;
1090 	dprintk("RPC:       xs_tcp_state_change client %p...\n", xprt);
1091 	dprintk("RPC:       state %x conn %d dead %d zapped %d\n",
1092 			sk->sk_state, xprt_connected(xprt),
1093 			sock_flag(sk, SOCK_DEAD),
1094 			sock_flag(sk, SOCK_ZAPPED));
1095 
1096 	switch (sk->sk_state) {
1097 	case TCP_ESTABLISHED:
1098 		spin_lock_bh(&xprt->transport_lock);
1099 		if (!xprt_test_and_set_connected(xprt)) {
1100 			struct sock_xprt *transport = container_of(xprt,
1101 					struct sock_xprt, xprt);
1102 
1103 			/* Reset TCP record info */
1104 			transport->tcp_offset = 0;
1105 			transport->tcp_reclen = 0;
1106 			transport->tcp_copied = 0;
1107 			transport->tcp_flags =
1108 				TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
1109 
1110 			xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1111 			xprt_wake_pending_tasks(xprt, 0);
1112 		}
1113 		spin_unlock_bh(&xprt->transport_lock);
1114 		break;
1115 	case TCP_SYN_SENT:
1116 	case TCP_SYN_RECV:
1117 		break;
1118 	case TCP_CLOSE_WAIT:
1119 		/* Try to schedule an autoclose RPC calls */
1120 		set_bit(XPRT_CLOSE_WAIT, &xprt->state);
1121 		if (test_and_set_bit(XPRT_LOCKED, &xprt->state) == 0)
1122 			queue_work(rpciod_workqueue, &xprt->task_cleanup);
1123 	default:
1124 		xprt_disconnect(xprt);
1125 	}
1126  out:
1127 	read_unlock(&sk->sk_callback_lock);
1128 }
1129 
1130 /**
1131  * xs_udp_write_space - callback invoked when socket buffer space
1132  *                             becomes available
1133  * @sk: socket whose state has changed
1134  *
1135  * Called when more output buffer space is available for this socket.
1136  * We try not to wake our writers until they can make "significant"
1137  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1138  * with a bunch of small requests.
1139  */
1140 static void xs_udp_write_space(struct sock *sk)
1141 {
1142 	read_lock(&sk->sk_callback_lock);
1143 
1144 	/* from net/core/sock.c:sock_def_write_space */
1145 	if (sock_writeable(sk)) {
1146 		struct socket *sock;
1147 		struct rpc_xprt *xprt;
1148 
1149 		if (unlikely(!(sock = sk->sk_socket)))
1150 			goto out;
1151 		if (unlikely(!(xprt = xprt_from_sock(sk))))
1152 			goto out;
1153 		if (unlikely(!test_and_clear_bit(SOCK_NOSPACE, &sock->flags)))
1154 			goto out;
1155 
1156 		xprt_write_space(xprt);
1157 	}
1158 
1159  out:
1160 	read_unlock(&sk->sk_callback_lock);
1161 }
1162 
1163 /**
1164  * xs_tcp_write_space - callback invoked when socket buffer space
1165  *                             becomes available
1166  * @sk: socket whose state has changed
1167  *
1168  * Called when more output buffer space is available for this socket.
1169  * We try not to wake our writers until they can make "significant"
1170  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1171  * with a bunch of small requests.
1172  */
1173 static void xs_tcp_write_space(struct sock *sk)
1174 {
1175 	read_lock(&sk->sk_callback_lock);
1176 
1177 	/* from net/core/stream.c:sk_stream_write_space */
1178 	if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) {
1179 		struct socket *sock;
1180 		struct rpc_xprt *xprt;
1181 
1182 		if (unlikely(!(sock = sk->sk_socket)))
1183 			goto out;
1184 		if (unlikely(!(xprt = xprt_from_sock(sk))))
1185 			goto out;
1186 		if (unlikely(!test_and_clear_bit(SOCK_NOSPACE, &sock->flags)))
1187 			goto out;
1188 
1189 		xprt_write_space(xprt);
1190 	}
1191 
1192  out:
1193 	read_unlock(&sk->sk_callback_lock);
1194 }
1195 
1196 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1197 {
1198 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1199 	struct sock *sk = transport->inet;
1200 
1201 	if (transport->rcvsize) {
1202 		sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1203 		sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1204 	}
1205 	if (transport->sndsize) {
1206 		sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1207 		sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1208 		sk->sk_write_space(sk);
1209 	}
1210 }
1211 
1212 /**
1213  * xs_udp_set_buffer_size - set send and receive limits
1214  * @xprt: generic transport
1215  * @sndsize: requested size of send buffer, in bytes
1216  * @rcvsize: requested size of receive buffer, in bytes
1217  *
1218  * Set socket send and receive buffer size limits.
1219  */
1220 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1221 {
1222 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1223 
1224 	transport->sndsize = 0;
1225 	if (sndsize)
1226 		transport->sndsize = sndsize + 1024;
1227 	transport->rcvsize = 0;
1228 	if (rcvsize)
1229 		transport->rcvsize = rcvsize + 1024;
1230 
1231 	xs_udp_do_set_buffer_size(xprt);
1232 }
1233 
1234 /**
1235  * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1236  * @task: task that timed out
1237  *
1238  * Adjust the congestion window after a retransmit timeout has occurred.
1239  */
1240 static void xs_udp_timer(struct rpc_task *task)
1241 {
1242 	xprt_adjust_cwnd(task, -ETIMEDOUT);
1243 }
1244 
1245 static unsigned short xs_get_random_port(void)
1246 {
1247 	unsigned short range = xprt_max_resvport - xprt_min_resvport;
1248 	unsigned short rand = (unsigned short) net_random() % range;
1249 	return rand + xprt_min_resvport;
1250 }
1251 
1252 /**
1253  * xs_set_port - reset the port number in the remote endpoint address
1254  * @xprt: generic transport
1255  * @port: new port number
1256  *
1257  */
1258 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1259 {
1260 	struct sockaddr *addr = xs_addr(xprt);
1261 
1262 	dprintk("RPC:       setting port for xprt %p to %u\n", xprt, port);
1263 
1264 	switch (addr->sa_family) {
1265 	case AF_INET:
1266 		((struct sockaddr_in *)addr)->sin_port = htons(port);
1267 		break;
1268 	case AF_INET6:
1269 		((struct sockaddr_in6 *)addr)->sin6_port = htons(port);
1270 		break;
1271 	default:
1272 		BUG();
1273 	}
1274 }
1275 
1276 static int xs_bind4(struct sock_xprt *transport, struct socket *sock)
1277 {
1278 	struct sockaddr_in myaddr = {
1279 		.sin_family = AF_INET,
1280 	};
1281 	struct sockaddr_in *sa;
1282 	int err;
1283 	unsigned short port = transport->port;
1284 
1285 	if (!transport->xprt.resvport)
1286 		port = 0;
1287 	sa = (struct sockaddr_in *)&transport->addr;
1288 	myaddr.sin_addr = sa->sin_addr;
1289 	do {
1290 		myaddr.sin_port = htons(port);
1291 		err = kernel_bind(sock, (struct sockaddr *) &myaddr,
1292 						sizeof(myaddr));
1293 		if (!transport->xprt.resvport)
1294 			break;
1295 		if (err == 0) {
1296 			transport->port = port;
1297 			break;
1298 		}
1299 		if (port <= xprt_min_resvport)
1300 			port = xprt_max_resvport;
1301 		else
1302 			port--;
1303 	} while (err == -EADDRINUSE && port != transport->port);
1304 	dprintk("RPC:       %s "NIPQUAD_FMT":%u: %s (%d)\n",
1305 			__FUNCTION__, NIPQUAD(myaddr.sin_addr),
1306 			port, err ? "failed" : "ok", err);
1307 	return err;
1308 }
1309 
1310 static int xs_bind6(struct sock_xprt *transport, struct socket *sock)
1311 {
1312 	struct sockaddr_in6 myaddr = {
1313 		.sin6_family = AF_INET6,
1314 	};
1315 	struct sockaddr_in6 *sa;
1316 	int err;
1317 	unsigned short port = transport->port;
1318 
1319 	if (!transport->xprt.resvport)
1320 		port = 0;
1321 	sa = (struct sockaddr_in6 *)&transport->addr;
1322 	myaddr.sin6_addr = sa->sin6_addr;
1323 	do {
1324 		myaddr.sin6_port = htons(port);
1325 		err = kernel_bind(sock, (struct sockaddr *) &myaddr,
1326 						sizeof(myaddr));
1327 		if (!transport->xprt.resvport)
1328 			break;
1329 		if (err == 0) {
1330 			transport->port = port;
1331 			break;
1332 		}
1333 		if (port <= xprt_min_resvport)
1334 			port = xprt_max_resvport;
1335 		else
1336 			port--;
1337 	} while (err == -EADDRINUSE && port != transport->port);
1338 	dprintk("RPC:       xs_bind6 "NIP6_FMT":%u: %s (%d)\n",
1339 		NIP6(myaddr.sin6_addr), port, err ? "failed" : "ok", err);
1340 	return err;
1341 }
1342 
1343 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1344 static struct lock_class_key xs_key[2];
1345 static struct lock_class_key xs_slock_key[2];
1346 
1347 static inline void xs_reclassify_socket4(struct socket *sock)
1348 {
1349 	struct sock *sk = sock->sk;
1350 
1351 	BUG_ON(sk->sk_lock.owner != NULL);
1352 	sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1353 		&xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1354 }
1355 
1356 static inline void xs_reclassify_socket6(struct socket *sock)
1357 {
1358 	struct sock *sk = sock->sk;
1359 
1360 	BUG_ON(sk->sk_lock.owner != NULL);
1361 	sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1362 		&xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1363 }
1364 #else
1365 static inline void xs_reclassify_socket4(struct socket *sock)
1366 {
1367 }
1368 
1369 static inline void xs_reclassify_socket6(struct socket *sock)
1370 {
1371 }
1372 #endif
1373 
1374 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
1375 {
1376 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1377 
1378 	if (!transport->inet) {
1379 		struct sock *sk = sock->sk;
1380 
1381 		write_lock_bh(&sk->sk_callback_lock);
1382 
1383 		sk->sk_user_data = xprt;
1384 		transport->old_data_ready = sk->sk_data_ready;
1385 		transport->old_state_change = sk->sk_state_change;
1386 		transport->old_write_space = sk->sk_write_space;
1387 		sk->sk_data_ready = xs_udp_data_ready;
1388 		sk->sk_write_space = xs_udp_write_space;
1389 		sk->sk_no_check = UDP_CSUM_NORCV;
1390 		sk->sk_allocation = GFP_ATOMIC;
1391 
1392 		xprt_set_connected(xprt);
1393 
1394 		/* Reset to new socket */
1395 		transport->sock = sock;
1396 		transport->inet = sk;
1397 
1398 		write_unlock_bh(&sk->sk_callback_lock);
1399 	}
1400 	xs_udp_do_set_buffer_size(xprt);
1401 }
1402 
1403 /**
1404  * xs_udp_connect_worker4 - set up a UDP socket
1405  * @work: RPC transport to connect
1406  *
1407  * Invoked by a work queue tasklet.
1408  */
1409 static void xs_udp_connect_worker4(struct work_struct *work)
1410 {
1411 	struct sock_xprt *transport =
1412 		container_of(work, struct sock_xprt, connect_worker.work);
1413 	struct rpc_xprt *xprt = &transport->xprt;
1414 	struct socket *sock = transport->sock;
1415 	int err, status = -EIO;
1416 
1417 	if (xprt->shutdown || !xprt_bound(xprt))
1418 		goto out;
1419 
1420 	/* Start by resetting any existing state */
1421 	xs_close(xprt);
1422 
1423 	if ((err = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock)) < 0) {
1424 		dprintk("RPC:       can't create UDP transport socket (%d).\n", -err);
1425 		goto out;
1426 	}
1427 	xs_reclassify_socket4(sock);
1428 
1429 	if (xs_bind4(transport, sock)) {
1430 		sock_release(sock);
1431 		goto out;
1432 	}
1433 
1434 	dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1435 			xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1436 
1437 	xs_udp_finish_connecting(xprt, sock);
1438 	status = 0;
1439 out:
1440 	xprt_wake_pending_tasks(xprt, status);
1441 	xprt_clear_connecting(xprt);
1442 }
1443 
1444 /**
1445  * xs_udp_connect_worker6 - set up a UDP socket
1446  * @work: RPC transport to connect
1447  *
1448  * Invoked by a work queue tasklet.
1449  */
1450 static void xs_udp_connect_worker6(struct work_struct *work)
1451 {
1452 	struct sock_xprt *transport =
1453 		container_of(work, struct sock_xprt, connect_worker.work);
1454 	struct rpc_xprt *xprt = &transport->xprt;
1455 	struct socket *sock = transport->sock;
1456 	int err, status = -EIO;
1457 
1458 	if (xprt->shutdown || !xprt_bound(xprt))
1459 		goto out;
1460 
1461 	/* Start by resetting any existing state */
1462 	xs_close(xprt);
1463 
1464 	if ((err = sock_create_kern(PF_INET6, SOCK_DGRAM, IPPROTO_UDP, &sock)) < 0) {
1465 		dprintk("RPC:       can't create UDP transport socket (%d).\n", -err);
1466 		goto out;
1467 	}
1468 	xs_reclassify_socket6(sock);
1469 
1470 	if (xs_bind6(transport, sock) < 0) {
1471 		sock_release(sock);
1472 		goto out;
1473 	}
1474 
1475 	dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1476 			xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1477 
1478 	xs_udp_finish_connecting(xprt, sock);
1479 	status = 0;
1480 out:
1481 	xprt_wake_pending_tasks(xprt, status);
1482 	xprt_clear_connecting(xprt);
1483 }
1484 
1485 /*
1486  * We need to preserve the port number so the reply cache on the server can
1487  * find our cached RPC replies when we get around to reconnecting.
1488  */
1489 static void xs_tcp_reuse_connection(struct rpc_xprt *xprt)
1490 {
1491 	int result;
1492 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1493 	struct sockaddr any;
1494 
1495 	dprintk("RPC:       disconnecting xprt %p to reuse port\n", xprt);
1496 
1497 	/*
1498 	 * Disconnect the transport socket by doing a connect operation
1499 	 * with AF_UNSPEC.  This should return immediately...
1500 	 */
1501 	memset(&any, 0, sizeof(any));
1502 	any.sa_family = AF_UNSPEC;
1503 	result = kernel_connect(transport->sock, &any, sizeof(any), 0);
1504 	if (result)
1505 		dprintk("RPC:       AF_UNSPEC connect return code %d\n",
1506 				result);
1507 }
1508 
1509 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
1510 {
1511 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1512 
1513 	if (!transport->inet) {
1514 		struct sock *sk = sock->sk;
1515 
1516 		write_lock_bh(&sk->sk_callback_lock);
1517 
1518 		sk->sk_user_data = xprt;
1519 		transport->old_data_ready = sk->sk_data_ready;
1520 		transport->old_state_change = sk->sk_state_change;
1521 		transport->old_write_space = sk->sk_write_space;
1522 		sk->sk_data_ready = xs_tcp_data_ready;
1523 		sk->sk_state_change = xs_tcp_state_change;
1524 		sk->sk_write_space = xs_tcp_write_space;
1525 		sk->sk_allocation = GFP_ATOMIC;
1526 
1527 		/* socket options */
1528 		sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
1529 		sock_reset_flag(sk, SOCK_LINGER);
1530 		tcp_sk(sk)->linger2 = 0;
1531 		tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1532 
1533 		xprt_clear_connected(xprt);
1534 
1535 		/* Reset to new socket */
1536 		transport->sock = sock;
1537 		transport->inet = sk;
1538 
1539 		write_unlock_bh(&sk->sk_callback_lock);
1540 	}
1541 
1542 	/* Tell the socket layer to start connecting... */
1543 	xprt->stat.connect_count++;
1544 	xprt->stat.connect_start = jiffies;
1545 	return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
1546 }
1547 
1548 /**
1549  * xs_tcp_connect_worker4 - connect a TCP socket to a remote endpoint
1550  * @work: RPC transport to connect
1551  *
1552  * Invoked by a work queue tasklet.
1553  */
1554 static void xs_tcp_connect_worker4(struct work_struct *work)
1555 {
1556 	struct sock_xprt *transport =
1557 		container_of(work, struct sock_xprt, connect_worker.work);
1558 	struct rpc_xprt *xprt = &transport->xprt;
1559 	struct socket *sock = transport->sock;
1560 	int err, status = -EIO;
1561 
1562 	if (xprt->shutdown || !xprt_bound(xprt))
1563 		goto out;
1564 
1565 	if (!sock) {
1566 		/* start from scratch */
1567 		if ((err = sock_create_kern(PF_INET, SOCK_STREAM, IPPROTO_TCP, &sock)) < 0) {
1568 			dprintk("RPC:       can't create TCP transport socket (%d).\n", -err);
1569 			goto out;
1570 		}
1571 		xs_reclassify_socket4(sock);
1572 
1573 		if (xs_bind4(transport, sock) < 0) {
1574 			sock_release(sock);
1575 			goto out;
1576 		}
1577 	} else
1578 		/* "close" the socket, preserving the local port */
1579 		xs_tcp_reuse_connection(xprt);
1580 
1581 	dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1582 			xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1583 
1584 	status = xs_tcp_finish_connecting(xprt, sock);
1585 	dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
1586 			xprt, -status, xprt_connected(xprt),
1587 			sock->sk->sk_state);
1588 	if (status < 0) {
1589 		switch (status) {
1590 			case -EINPROGRESS:
1591 			case -EALREADY:
1592 				goto out_clear;
1593 			case -ECONNREFUSED:
1594 			case -ECONNRESET:
1595 				/* retry with existing socket, after a delay */
1596 				break;
1597 			default:
1598 				/* get rid of existing socket, and retry */
1599 				xs_close(xprt);
1600 				break;
1601 		}
1602 	}
1603 out:
1604 	xprt_wake_pending_tasks(xprt, status);
1605 out_clear:
1606 	xprt_clear_connecting(xprt);
1607 }
1608 
1609 /**
1610  * xs_tcp_connect_worker6 - connect a TCP socket to a remote endpoint
1611  * @work: RPC transport to connect
1612  *
1613  * Invoked by a work queue tasklet.
1614  */
1615 static void xs_tcp_connect_worker6(struct work_struct *work)
1616 {
1617 	struct sock_xprt *transport =
1618 		container_of(work, struct sock_xprt, connect_worker.work);
1619 	struct rpc_xprt *xprt = &transport->xprt;
1620 	struct socket *sock = transport->sock;
1621 	int err, status = -EIO;
1622 
1623 	if (xprt->shutdown || !xprt_bound(xprt))
1624 		goto out;
1625 
1626 	if (!sock) {
1627 		/* start from scratch */
1628 		if ((err = sock_create_kern(PF_INET6, SOCK_STREAM, IPPROTO_TCP, &sock)) < 0) {
1629 			dprintk("RPC:       can't create TCP transport socket (%d).\n", -err);
1630 			goto out;
1631 		}
1632 		xs_reclassify_socket6(sock);
1633 
1634 		if (xs_bind6(transport, sock) < 0) {
1635 			sock_release(sock);
1636 			goto out;
1637 		}
1638 	} else
1639 		/* "close" the socket, preserving the local port */
1640 		xs_tcp_reuse_connection(xprt);
1641 
1642 	dprintk("RPC:       worker connecting xprt %p to address: %s\n",
1643 			xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1644 
1645 	status = xs_tcp_finish_connecting(xprt, sock);
1646 	dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
1647 			xprt, -status, xprt_connected(xprt), sock->sk->sk_state);
1648 	if (status < 0) {
1649 		switch (status) {
1650 			case -EINPROGRESS:
1651 			case -EALREADY:
1652 				goto out_clear;
1653 			case -ECONNREFUSED:
1654 			case -ECONNRESET:
1655 				/* retry with existing socket, after a delay */
1656 				break;
1657 			default:
1658 				/* get rid of existing socket, and retry */
1659 				xs_close(xprt);
1660 				break;
1661 		}
1662 	}
1663 out:
1664 	xprt_wake_pending_tasks(xprt, status);
1665 out_clear:
1666 	xprt_clear_connecting(xprt);
1667 }
1668 
1669 /**
1670  * xs_connect - connect a socket to a remote endpoint
1671  * @task: address of RPC task that manages state of connect request
1672  *
1673  * TCP: If the remote end dropped the connection, delay reconnecting.
1674  *
1675  * UDP socket connects are synchronous, but we use a work queue anyway
1676  * to guarantee that even unprivileged user processes can set up a
1677  * socket on a privileged port.
1678  *
1679  * If a UDP socket connect fails, the delay behavior here prevents
1680  * retry floods (hard mounts).
1681  */
1682 static void xs_connect(struct rpc_task *task)
1683 {
1684 	struct rpc_xprt *xprt = task->tk_xprt;
1685 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1686 
1687 	if (xprt_test_and_set_connecting(xprt))
1688 		return;
1689 
1690 	if (transport->sock != NULL) {
1691 		dprintk("RPC:       xs_connect delayed xprt %p for %lu "
1692 				"seconds\n",
1693 				xprt, xprt->reestablish_timeout / HZ);
1694 		queue_delayed_work(rpciod_workqueue,
1695 				   &transport->connect_worker,
1696 				   xprt->reestablish_timeout);
1697 		xprt->reestablish_timeout <<= 1;
1698 		if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
1699 			xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
1700 	} else {
1701 		dprintk("RPC:       xs_connect scheduled xprt %p\n", xprt);
1702 		queue_delayed_work(rpciod_workqueue,
1703 				   &transport->connect_worker, 0);
1704 	}
1705 }
1706 
1707 /**
1708  * xs_udp_print_stats - display UDP socket-specifc stats
1709  * @xprt: rpc_xprt struct containing statistics
1710  * @seq: output file
1711  *
1712  */
1713 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1714 {
1715 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1716 
1717 	seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %Lu %Lu\n",
1718 			transport->port,
1719 			xprt->stat.bind_count,
1720 			xprt->stat.sends,
1721 			xprt->stat.recvs,
1722 			xprt->stat.bad_xids,
1723 			xprt->stat.req_u,
1724 			xprt->stat.bklog_u);
1725 }
1726 
1727 /**
1728  * xs_tcp_print_stats - display TCP socket-specifc stats
1729  * @xprt: rpc_xprt struct containing statistics
1730  * @seq: output file
1731  *
1732  */
1733 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1734 {
1735 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1736 	long idle_time = 0;
1737 
1738 	if (xprt_connected(xprt))
1739 		idle_time = (long)(jiffies - xprt->last_used) / HZ;
1740 
1741 	seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu %Lu %Lu\n",
1742 			transport->port,
1743 			xprt->stat.bind_count,
1744 			xprt->stat.connect_count,
1745 			xprt->stat.connect_time,
1746 			idle_time,
1747 			xprt->stat.sends,
1748 			xprt->stat.recvs,
1749 			xprt->stat.bad_xids,
1750 			xprt->stat.req_u,
1751 			xprt->stat.bklog_u);
1752 }
1753 
1754 static struct rpc_xprt_ops xs_udp_ops = {
1755 	.set_buffer_size	= xs_udp_set_buffer_size,
1756 	.reserve_xprt		= xprt_reserve_xprt_cong,
1757 	.release_xprt		= xprt_release_xprt_cong,
1758 	.rpcbind		= rpcb_getport_async,
1759 	.set_port		= xs_set_port,
1760 	.connect		= xs_connect,
1761 	.buf_alloc		= rpc_malloc,
1762 	.buf_free		= rpc_free,
1763 	.send_request		= xs_udp_send_request,
1764 	.set_retrans_timeout	= xprt_set_retrans_timeout_rtt,
1765 	.timer			= xs_udp_timer,
1766 	.release_request	= xprt_release_rqst_cong,
1767 	.close			= xs_close,
1768 	.destroy		= xs_destroy,
1769 	.print_stats		= xs_udp_print_stats,
1770 };
1771 
1772 static struct rpc_xprt_ops xs_tcp_ops = {
1773 	.reserve_xprt		= xprt_reserve_xprt,
1774 	.release_xprt		= xs_tcp_release_xprt,
1775 	.rpcbind		= rpcb_getport_async,
1776 	.set_port		= xs_set_port,
1777 	.connect		= xs_connect,
1778 	.buf_alloc		= rpc_malloc,
1779 	.buf_free		= rpc_free,
1780 	.send_request		= xs_tcp_send_request,
1781 	.set_retrans_timeout	= xprt_set_retrans_timeout_def,
1782 	.close			= xs_close,
1783 	.destroy		= xs_destroy,
1784 	.print_stats		= xs_tcp_print_stats,
1785 };
1786 
1787 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
1788 				      unsigned int slot_table_size)
1789 {
1790 	struct rpc_xprt *xprt;
1791 	struct sock_xprt *new;
1792 
1793 	if (args->addrlen > sizeof(xprt->addr)) {
1794 		dprintk("RPC:       xs_setup_xprt: address too large\n");
1795 		return ERR_PTR(-EBADF);
1796 	}
1797 
1798 	new = kzalloc(sizeof(*new), GFP_KERNEL);
1799 	if (new == NULL) {
1800 		dprintk("RPC:       xs_setup_xprt: couldn't allocate "
1801 				"rpc_xprt\n");
1802 		return ERR_PTR(-ENOMEM);
1803 	}
1804 	xprt = &new->xprt;
1805 
1806 	xprt->max_reqs = slot_table_size;
1807 	xprt->slot = kcalloc(xprt->max_reqs, sizeof(struct rpc_rqst), GFP_KERNEL);
1808 	if (xprt->slot == NULL) {
1809 		kfree(xprt);
1810 		dprintk("RPC:       xs_setup_xprt: couldn't allocate slot "
1811 				"table\n");
1812 		return ERR_PTR(-ENOMEM);
1813 	}
1814 
1815 	memcpy(&xprt->addr, args->dstaddr, args->addrlen);
1816 	xprt->addrlen = args->addrlen;
1817 	if (args->srcaddr)
1818 		memcpy(&new->addr, args->srcaddr, args->addrlen);
1819 	new->port = xs_get_random_port();
1820 
1821 	return xprt;
1822 }
1823 
1824 /**
1825  * xs_setup_udp - Set up transport to use a UDP socket
1826  * @args: rpc transport creation arguments
1827  *
1828  */
1829 struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
1830 {
1831 	struct sockaddr *addr = args->dstaddr;
1832 	struct rpc_xprt *xprt;
1833 	struct sock_xprt *transport;
1834 
1835 	xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries);
1836 	if (IS_ERR(xprt))
1837 		return xprt;
1838 	transport = container_of(xprt, struct sock_xprt, xprt);
1839 
1840 	xprt->prot = IPPROTO_UDP;
1841 	xprt->tsh_size = 0;
1842 	/* XXX: header size can vary due to auth type, IPv6, etc. */
1843 	xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
1844 
1845 	xprt->bind_timeout = XS_BIND_TO;
1846 	xprt->connect_timeout = XS_UDP_CONN_TO;
1847 	xprt->reestablish_timeout = XS_UDP_REEST_TO;
1848 	xprt->idle_timeout = XS_IDLE_DISC_TO;
1849 
1850 	xprt->ops = &xs_udp_ops;
1851 
1852 	if (args->timeout)
1853 		xprt->timeout = *args->timeout;
1854 	else
1855 		xprt_set_timeout(&xprt->timeout, 5, 5 * HZ);
1856 
1857 	switch (addr->sa_family) {
1858 	case AF_INET:
1859 		if (((struct sockaddr_in *)addr)->sin_port != htons(0))
1860 			xprt_set_bound(xprt);
1861 
1862 		INIT_DELAYED_WORK(&transport->connect_worker,
1863 					xs_udp_connect_worker4);
1864 		xs_format_ipv4_peer_addresses(xprt);
1865 		break;
1866 	case AF_INET6:
1867 		if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
1868 			xprt_set_bound(xprt);
1869 
1870 		INIT_DELAYED_WORK(&transport->connect_worker,
1871 					xs_udp_connect_worker6);
1872 		xs_format_ipv6_peer_addresses(xprt);
1873 		break;
1874 	default:
1875 		kfree(xprt);
1876 		return ERR_PTR(-EAFNOSUPPORT);
1877 	}
1878 
1879 	dprintk("RPC:       set up transport to address %s\n",
1880 			xprt->address_strings[RPC_DISPLAY_ALL]);
1881 
1882 	if (try_module_get(THIS_MODULE))
1883 		return xprt;
1884 
1885 	kfree(xprt->slot);
1886 	kfree(xprt);
1887 	return ERR_PTR(-EINVAL);
1888 }
1889 
1890 /**
1891  * xs_setup_tcp - Set up transport to use a TCP socket
1892  * @args: rpc transport creation arguments
1893  *
1894  */
1895 struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
1896 {
1897 	struct sockaddr *addr = args->dstaddr;
1898 	struct rpc_xprt *xprt;
1899 	struct sock_xprt *transport;
1900 
1901 	xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries);
1902 	if (IS_ERR(xprt))
1903 		return xprt;
1904 	transport = container_of(xprt, struct sock_xprt, xprt);
1905 
1906 	xprt->prot = IPPROTO_TCP;
1907 	xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
1908 	xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
1909 
1910 	xprt->bind_timeout = XS_BIND_TO;
1911 	xprt->connect_timeout = XS_TCP_CONN_TO;
1912 	xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1913 	xprt->idle_timeout = XS_IDLE_DISC_TO;
1914 
1915 	xprt->ops = &xs_tcp_ops;
1916 
1917 	if (args->timeout)
1918 		xprt->timeout = *args->timeout;
1919 	else
1920 		xprt_set_timeout(&xprt->timeout, 2, 60 * HZ);
1921 
1922 	switch (addr->sa_family) {
1923 	case AF_INET:
1924 		if (((struct sockaddr_in *)addr)->sin_port != htons(0))
1925 			xprt_set_bound(xprt);
1926 
1927 		INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_connect_worker4);
1928 		xs_format_ipv4_peer_addresses(xprt);
1929 		break;
1930 	case AF_INET6:
1931 		if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
1932 			xprt_set_bound(xprt);
1933 
1934 		INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_connect_worker6);
1935 		xs_format_ipv6_peer_addresses(xprt);
1936 		break;
1937 	default:
1938 		kfree(xprt);
1939 		return ERR_PTR(-EAFNOSUPPORT);
1940 	}
1941 
1942 	dprintk("RPC:       set up transport to address %s\n",
1943 			xprt->address_strings[RPC_DISPLAY_ALL]);
1944 
1945 	if (try_module_get(THIS_MODULE))
1946 		return xprt;
1947 
1948 	kfree(xprt->slot);
1949 	kfree(xprt);
1950 	return ERR_PTR(-EINVAL);
1951 }
1952 
1953 static struct xprt_class	xs_udp_transport = {
1954 	.list		= LIST_HEAD_INIT(xs_udp_transport.list),
1955 	.name		= "udp",
1956 	.owner		= THIS_MODULE,
1957 	.family		= AF_INET,
1958 	.protocol	= IPPROTO_UDP,
1959 	.setup		= xs_setup_udp,
1960 };
1961 
1962 static struct xprt_class	xs_tcp_transport = {
1963 	.list		= LIST_HEAD_INIT(xs_tcp_transport.list),
1964 	.name		= "tcp",
1965 	.owner		= THIS_MODULE,
1966 	.family		= AF_INET,
1967 	.protocol	= IPPROTO_TCP,
1968 	.setup		= xs_setup_tcp,
1969 };
1970 
1971 /**
1972  * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
1973  *
1974  */
1975 int init_socket_xprt(void)
1976 {
1977 #ifdef RPC_DEBUG
1978 	if (!sunrpc_table_header)
1979 		sunrpc_table_header = register_sysctl_table(sunrpc_table);
1980 #endif
1981 
1982 	xprt_register_transport(&xs_udp_transport);
1983 	xprt_register_transport(&xs_tcp_transport);
1984 
1985 	return 0;
1986 }
1987 
1988 /**
1989  * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
1990  *
1991  */
1992 void cleanup_socket_xprt(void)
1993 {
1994 #ifdef RPC_DEBUG
1995 	if (sunrpc_table_header) {
1996 		unregister_sysctl_table(sunrpc_table_header);
1997 		sunrpc_table_header = NULL;
1998 	}
1999 #endif
2000 
2001 	xprt_unregister_transport(&xs_udp_transport);
2002 	xprt_unregister_transport(&xs_tcp_transport);
2003 }
2004